Premium
Quality of first trimester risk prediction models for pre‐eclampsia: a systematic review
Author(s) -
Brunelli VB,
Prefumo F
Publication year - 2015
Publication title -
bjog: an international journal of obstetrics and gynaecology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.157
H-Index - 164
eISSN - 1471-0528
pISSN - 1470-0328
DOI - 10.1111/1471-0528.13334
Subject(s) - eclampsia , medicine , interquartile range , logistic regression , overfitting , sample size determination , predictive modelling , statistics , pregnancy , computer science , machine learning , mathematics , genetics , artificial neural network , biology
Background There is an increasing interest in first trimester risk prediction models for pre‐eclampsia. Objectives To systematically review and critically assess the building and reporting of methods used to develop first trimester risk prediction models for pre‐eclampsia. Search strategy Search of PubMed and EMBASE databases from inception to July 2013. Selection criteria Logistic regression model for predicting the risk of pre‐eclampsia in the first trimester, including uterine artery Doppler among independent variables. Data collection and analysis We extracted information on study design, outcome definition, participant recruitment, sample size and number of events, risk predictors and their selection and treatment, model‐building strategies, missing data, overfitting and validation. Main results The initial search identified 80 articles. A total of 24 studies were eligible for review, from which 38 predictive models were identified. The median number of study participants was 697 [interquartile range (IQR) 377– 5126]. The median number of cases of pre‐eclampsia per model was 37 (IQR 19–97). The median number of risk predictors was 5 (IQR 3.75–7). In 22% of the models, the number of events per variable was fewer than the commonly recommended value of 10 events per predictor; this proportion increased to 94% in models for early pre‐eclampsia. Treatment and handling of missing data were not reported in 37 models. Only three models reported model validation. Conclusions We found frequent methodological deficiencies in studies reporting risk prediction models for pre‐eclampsia. This may limit their reliability and validity.